Project Overview
- I worked on a set of interlinked modules which work together to provide diagnostic support to respiratory diseases
- The support on medical reports involved negation detection using semantic rules to find out the presence or absence of a medical concept and NLP modeling for summarization of the identified concepts
- This also involved having lucene indices for the text, topic identification and boundary detection for efficient reporting
- The computer vision part was looking to predict anomalous areas compared to a healthy scan, which can be found in cases such as pulmonary embolism. This part was more of a back up support than full ownership
- Other than the ML parts, there was also a fair amount of HIPAA training and PII sanity checks involved in this project
Skills
PII / sensitive data handling
The HIPAA training and framework for this project, put into practice for me the sensitivity and value of data and how using it is a privilege. And that privilege comes with responsibility of ensuring outcome out of it's use
Elastic indices
Since medical records can go back several years for many patients and take a long time to process, the use of lucene indices helped speed up processing to enable real time outcomes
Mission critical product metrics
Watson Health being a critical client facing project came with some high expectations in terms of the model metrics and performance. Unlike optimizing metrics just to look good on paper, this project involved optimizing them for mission critical real world use